In this series of presentations, Susan Athey
, Brian Knigh
t and Justin Rao
each examined different ways that social media usage affects opinion formation, voting patterns, and "filter bubbles" of biased information constructed by individual users. A concluding panel moderated by Joel Waldfogel synthesized some common insights on the nature of bias as it travels across social media.
In looking at a broad collection of different browsing data from internet users, Susan Athey and her coauthors considered how social media affects the type of news that people choose to consume, as well as the degree to which that media keeps them informed. More specific to her discussion at this conference, she examined how social media might affect a user's political biases.
Athey acknowledged that empirical research in this area presents a serious measurement challenge. Scholars must find a way to reliably characterize what people are reading, how they interact with it, and how they are referred from social network to news outlet and vice versa. And once they do, there’s the realities of users rapidly embracing new networks, changing the degree to which they interact with any one network, increasingly abandoning old news outlets for new ones. Athey used a combination of browser toolbar data, article data, a broad list of news outlets, Twitter’s “firehose” of data, and analytics from news outlet websites to piece together a relatively complete picture, but further study will require ever-adapting research approaches.
Several broad patterns emerged in the study. In general, users tended to find news within their own political sphere, based on the network of people with whom they choose to connect. More politically-engaged readers were referred from social media sources, mostly due to the fact that content shared by their insulating, self-created social networks tended to create a mix of topics that was more likely to inspire impassioned defenses, outrage or approval. Conservative readers see almost no liberal news, and likewise, liberal users hardly read any conservative news through social media. But this doesn't hold for all types of media. Broad topics like sports were so all encompassing in terms of how people responded to them that little could be gleaned from the data about how the media itself shifted opinions of users.
In Athey's view, this says that when it comes to social media, political bias operates more at the topic level than at the outlet level. Moreover, only a certain subset of those topics affect user polarization, with broader topics demonstrating little observable effect on users reading about them.
While Athey discussed how the mix of topics that a user selects can ultimately shift how they are affected by what they read, Brian Knight asked why that media bubble might form around a user in the first place. Namely, does social media expose us to more diverse opinions, or allow us to connect with like-minded individuals as method of walling ourselves off from differing viewpoints? The latter phenomena, called homophily, was the object of study by Knight and his coauthor
, who looked at how homophily develops among groups of Twitter users over time.
The goal: to construct a network of politically engaged Twitter users. To do so, Knight noted which politicians a user followed on Twitter (defined as candidates for the US House of Representatives in 2012) as a proxy for their ideology. Those 2.2 million users were sorted as 'Republican' or 'Democrat' if they followed more candidates of one party than another. These groups were then subdivided by candidate geography, allowing friendship tests from Currarini, Jackson, and Pin (2009); these tests allowed Knight to find larger friendship groups, more likely to contain homophily, and compare the emergence of homophily in real-world groups to similar social media counterparts. They then employed Gentzkow and Shapiro's (2011) model for measuring ideological segregation between groups, examining the media outlets that users choose to follow as a point of comparison to the politicians they select.
Knight says that the findings establish that homophily among Twitter users doesn't function all that differently from that in the real world. "When used as a social network, Twitter is highly segregated along ideological lines, placing it on a par with other face-to-face social interactions," says Knight. "When Twitter is used to follow media outlets, segregation is significantly lower."
Justin Rao discussed his work at Microsoft looking at similar issues surrounding homophily among social media users, labeling it a "filter bubble"
– an algorithmically-created filter around information in the form of the network of other users that surrounds each individual user. Rao and his coauthors sought to examine how this shift toward "choice technology" has affected the baseline of media being consumed over time, specifically how political polarization might shift how news is consumed and, in some cases, written.
Taking browsing records of 1.2 million de-identified users and comparing them to a "news corpus" of article text scraped from top news domains in an open directory project, they compared what people chose to read to the "conservative share" of the chosen news outlet, established by what fraction of its readership voted Republican in the 2012 presidential election.
The analysis yielded a few interesting insights: for one, people who sought out opinion tended to choose more polarized outlets, dominated by terms that subjectively shift based on your point of view, like "gun rights," "abortion," and "death panels." "You use a different language to describe a topic if you come from a conservative area versus a liberal area," said Rao.
But on a deeper level, Rao found that users employing direct navigation of news sites–going to NYTimes.com and clicking around–made for far less polarized users. Why? Rao hypothesizes that while search and social push people to the fringe of polarized discussion, users directly navigating the web regress to the mean online news agenda, meaning that they have their own, echo chamber of popular news content entirely separate from the smaller personalized bubbles that social media can create.
In the panel discussion that followed, the three researchers discussed their shared challenges in examining the impact of social media on news consumption, the greatest of which being a lack thereof. "I'm still mystified by how small [the effect] is," said Athey. Knight accounted for that small effect simply–people use social media for many, many things, and so even as more people use social media, the relative percentage using it for news consumption pales in comparison to other uses. Rao agreed, noting that search is substantially bigger than social when it comes to news discovery, but stressing that findings like that don't mean there isn't a subset of people for which social media are critical for news consumption.
Data presents a constant challenge, but social media researchers have even broader problem to solve when it comes to asking about the impact of social media on users. "We need unified framework for what people choose to consume, and what impact it has on outcomes," said Knight. Isolating the effects of any one network–Facebook, Twitter, or whatever comes next–can be difficult to impossible, since the wide reach of such networks can show up invisibly in a wide array of data sources; a complete picture of what affects users represents a constantly shifting target.